The usual prognosis for long-term disabled elders is that they are not likely to improve or regain independent functional status. But computed transition probabilities reveal significant probabilities of improvement at all levels of disability. Yet, not much is known about what determines who improves and who deteriorates. This study plans to identify the characteristics that differentiate between improving and deteriorating functional status and to estimate how much each characteristic contributes to the probabilities of improvement or deterioration. By finding these characteristics, policy makers will be better equipped to devise measures that raise the probability of improvement and lower the probability of deterioration. The study will be based on the National Long-Term Care Surveys (NLTCS) of 1982 and 1984 and the linked Medicare (Part A) file. Improvement and deterioration of functional status are measured by comparing limitations in the physical activities of daily living in both years. The level of disability in these years is known for 5,388 persons (896 improved, 1,826 remained the same, and 2,666 deteriorated). The regression analysis will be based on the universal logit model, which will provide separate coefficient vectors for "improved" and "deteriorated," using "same" as a basis for comparison. Partial derivatives or partial differences, which are computed from the regression coefficients (of 13 types of variables), can measure how much each of the specified explanatory variables contributes to changes in the probabilities of improvement and deterioration. The magnitudes of the effects of the specified variables can provide an order of priority for targeting the policy-amenable factors to enhance the likelihood of improvement and diminish the possibility of deterioration.